Object Detection Using Efficient Partitioning and Frame Reduction

Rapid object detection is crucial for safety-critical applications, such as post-event analysis of surveillance videos in crime investigations and the operation of autonomous vehicles. The objective of this paper is to improve the speed of object detection through efficient partitioning and frame reduction using parallel processing. Techniques to improve Spark's default partitioning by using entropy-based algorithms to create more evenly distributed partitions based on the estimated workload of the frames are considered. Redundant frames are removed to efficiently reduce the workload. Algorithms for removing redundant frames are introduced and evaluated for their effectiveness in comparison to random and fixed frame removals. The results from this paper demonstrate that partitioning algorithms that estimate workload using entropy provide faster processing time when compared to those that use random partitioning. The results also indicate that frame removal algorithms that use frame-specific information, like entropy difference between frames, further improve performance without notably reducing detection accuracy when compared to those that do not utilize frame-specific information.